A cautionary tale of side-by-side evaluations while developing emotional expression for intelligent virtual agents